AI: can risk be presented without steering the decision?
When managers ask AI to summarise a trade-off, they also delegate part of its framing. The precise question is whether presenting identical consequences as gains retained or losses incurred can change the option selected.
What the literature establishes
Amos Tversky and Daniel Kahneman (1981, Science) demonstrated experimentally that different descriptions of equivalent consequences can shift preferences between a certain option and a risky one. In their classic public-health problem, describing outcomes as lives saved or lives lost changes the choices observed. Accurate information alone therefore does not guarantee a stable decision.
Different frames, different effects
Irwin P. Levin, Sandra L. Schneider and Gary J. Gaeth (1998, Organizational Behavior and Human Decision Processes) distinguish risky-choice framing, attribute framing and goal framing. Describing a characteristic favourably does not involve the same mechanism as presenting a choice between certainty and risk. For managers, this distinction helps prevent every AI-generated reformulation from being treated as merely a matter of style.
The hasty conclusion
One might conclude that simply asking AI for “neutral” wording is enough. These studies demonstrate neither that such an instruction eliminates framing effects nor that AI systematically amplifies them. Presenting favourable and unfavourable consequences together is better understood as a precaution to test than as a guarantee of impartiality. (our executive and employee training programmes)
What the evidence does not allow us to generalise
These articles examine neither today’s generative assistants nor business decisions in Lugano. Real decisions also involve constraints, discussions and responsibilities that experimental problems simplify. Comparing two formulations is informative only if probabilities, amounts, time horizons and available options remain identical.
A practical check in Lugano
As part of the SHR programme “Managing in the Age of Artificial Intelligence”, an exercise involves preparing two Italian versions of the same fictional trade-off for managers in banking, fashion, trading or Italian-speaking family SMEs in Lugano: one centred on gains retained, the other on losses incurred. Once a human reviewer has checked their equivalence, randomly assign participants to the versions and ask them to choose individually before any discussion. Measure the percentage-point difference between the proportions choosing the risky option in each group, reporting group sizes and the reasons given alongside it. This difference provides a local signal to discuss, not general proof; the exercise ends by presenting gains and losses together and examining any changes in choice. To go further: explore the Managing in the Age of Artificial Intelligence training in Lugano in the canton of Ticino, or browse our executive and employee training programmes in Switzerland.
In pictures: Managing in the Age of Artificial Intelligence in Lugano
- managing in the age of artificial intelligence
- Lugano
- research
- training lugano